Carnegie Mellon University

38616 - Neural Networks and Deep Learning in Science

The course focuses on practice and applications of deep learning by exploring foundational concepts, structuring popular networks and implementing models through modern technologies (python, Jupyter notebooks and PyTorch). Other topics may include image recognition, machine translation, natural language processing, parallelism, GPU distributed computing, cloud technologies, inference and parameter fitting in deep networks. Course uses large datasets hosted by PSC.

Potential topics include:
  • Basic concepts: Model accuracy, prediction accuracy, interpretability, supervised and un- supervised training, regularization.
  • Artificial neural networks, feed-forward, activation functions, loss functions.
  • Non-linear optimization, gradient descent, back-propagation
  • Deep Learning tools: PyTorch, AWS cloud
  • Autoencoders, dense embedding, dimensionality reduction
  • Convolutional networks, transfer learning, applications in image processing and sciences
  • Recurrent networks, LSTM, GRU, applications in NLP
  • Other topics: GANs, Reinforcement Learning, Multitask Learning, advanced applications of deep learning in chemical and biological sciences.